# Health, client and member data

*At work · 2 of 6*

- What's fine to paste, and who decides, in a small practice.
- The incident clocks the people who handle incidents will care about.
- The questions to ask an AI vendor, and the list to keep.

5 min read

Checked 3 November 2026. This isn't legal advice, and the rules change, so check the sources linked under each point. The laws and regulators named here are Australian. Elsewhere the rules differ, and the same questions still apply. The first section is for a small practice or clinic, and the rest is for a regulated business.

## In a small practice

The Privacy Act covers health service providers whatever their size. The Office of the Australian Information Commissioner (OAIC) says that "Regardless of turnover, the Privacy Act covers any business that is" a health service provider, among others. Otherwise a small business is one with "an annual turnover of $3 million or less". So a four-person clinic is covered. Source: OAIC on small business (https://www.oaic.gov.au/privacy/privacy-for-organisations/small-business).

The same page says some small businesses "may be required to comply with" specific rules on top, including the Privacy (Tax File Number) Rule 2015 for tax file numbers. So a small bookkeeper or accountant who handles them has rules for those too.

- Fine to paste. A message with [patient] and [time] gaps and no health details, such as a missed-appointment text that says "Hi [patient], we missed you at your [time] appointment today".
- Not fine. Real names with an injury, a referral or a claim number, or any email or note you haven't scrubbed.
- Training and temporary chats. On a free or personal account, find the setting that lets the company train on your chats and turn it off. Real health details stay out of any chat. For scrubbed work, a temporary chat keeps it out of your history, though the company can still keep it for a short time. Workplace plans often don't train on your data by default, and the vendor's privacy page will say.
- Who decides. The practice owner decides whether AI is used with patient records, and which tool. If that's you, using a free account for jobs with gaps and no real details is your call, so decide and write it down. If it isn't you, ask the owner first.
- Don't connect it. Don't link the practice inbox or calendar to a personal account.
- Templates. Save a finished message in your booking system's or email's templates, fill in the real name by hand and read it before it goes.

## In a regulated business

Use your firm's approved tool and its own policies. The APRA standards below apply only to APRA-regulated entities, such as banks, insurers and super funds. Other regulated work, like accounting, law or financial advice, has its own professional and privacy duties, so ask your professional body or adviser which rules and clocks apply to you. Everything in Set the rules first (/guide/05-safety-and-governance.html) applies, and three more things matter at this scale: the incident process, the vendor and the list of what's running.

## When something goes wrong

Report to whoever handles security incidents before anyone changes or deletes anything, so the logs and evidence are kept. Then follow their steps. These clocks are pointers for them:

- Notifiable Data Breaches scheme. A covered organisation must "take all reasonable steps to complete the assessment within 30 calendar days" of suspecting an eligible data breach. It must give the Commissioner a statement "as soon as practicable", and "notify individuals as soon as practicable". An eligible breach is unauthorised access, disclosure or loss of personal information that's likely to result in serious harm, where remedial action hasn't prevented that risk. The OAIC says to treat the 30 days as a maximum. Source: OAIC on the NDB scheme (https://www.oaic.gov.au/privacy/notifiable-data-breaches/preventing-preparing-for-and-responding-to-data-breaches/data-breach-preparation-and-response/part-4-notifiable-data-breach-ndb-scheme).
- APRA's CPS 234, paragraph 35. An APRA-regulated entity "must notify APRA as soon as possible and, in any case, no later than 72 hours, after becoming aware of an information security incident" that materially affected, or could have materially affected, the entity or its customers, or that has been notified to other regulators. Source: APRA's CPS 234 (https://www.apra.gov.au/standards/cps-234).
- APRA's CPS 234, paragraph 36. The entity must notify APRA "no later than 10 business days, after it becomes aware of a material information security control weakness" that it doesn't expect to fix in time. Paragraph 16 adds that where a third party manages its information assets, it "must assess the information security capability of that party", and an AI vendor can be that party.
- APRA's CPS 230, Operational Risk Management. The current version commences on 1 July 2026. Material service providers are "those on which the entity relies to undertake a critical operation", and the entity sends APRA a register of them each year. Whether an AI vendor belongs on it is a question for whoever owns compliance. Source: APRA's CPS 230 (https://www.apra.gov.au/standards/cps-230).
- The OAIC on AI products. Its guidance says organisations "should take a 'privacy by design' approach, which includes conducting a Privacy Impact Assessment", and recommends that they "do not enter personal information, and particularly sensitive information, into publicly available generative AI tools". Source: OAIC guidance on commercially available AI products (https://www.oaic.gov.au/privacy/privacy-guidance-for-organisations-and-government-agencies/guidance-on-privacy-and-the-use-of-commercially-available-ai-products), published 21 October 2024 and updated 17 January 2025.

## Questions for the vendor

- Is our data used to train models?
- How long is it kept, and where, including outside Australia?
- Who are its sub-processors?
- Can we get logs of what the AI read and did?
- What does the vendor do, and tell us, after an incident?
- Can our admins restrict connectors and actions?
- What certifications does it hold, and do we get audit rights?
- How fast will it tell us about a breach, measured against our own 72-hour clock?

## An owner and a list

Every agent or routine has a named owner. Keep one list of what each is, what it can read and do, who approved it and when it was last reviewed. The connection log (/guide/kit/connection-log.md) is a ready-made shape, and Routines (/guide/11-put-it-on-a-schedule.html) covers keeping routines on it.

Write the approval path on one page. Who asks, who approves, what they check (the data buckets, the vendor's answers and the controls), and how often it's reviewed.

Then sort your protections into the two kinds in Set the rules first (/guide/05-safety-and-governance.html#s4). Prompt rules and instructions are requests that can fail. Missing permissions, read-only connections and admin restrictions are controls. A risk review should rest on the second kind.
